The Question Paper Two Couldn't Answer

Here is the question, in one line: is the disagreement between two diagonals equal to the disagreement between two true observer entropies?

It sounds like a technicality. It is the whole programme.

The true observer entropy is the von Neumann entropy of the actual reduced state $\rho_{R_X}$ – a full quantum object, built from the random code $V$, with off-diagonal coherences, with a fluctuating normalization, with all the structure that makes quantum entropy hard. The thing we know how to compute is $H(P_X)$, the Shannon entropy of the diagonal $P_X$ – an ordinary classical quantity, a number you get from a bar chart. Paper II proved that on average the state collapses to the diagonal. But entropy is not linear, and the entropy of an average is not the average of an entropy. The average collapsing to the diagonal does not, by itself, mean the entropy is governed by the diagonal.

So we had two quantities, and a gap between them: $\Delta \;=\; \big[S(\rho_{R_A}) - S(\rho_{R_B})\big] \;-\; \big[H(P_A) - H(P_B)\big].$ If $\Delta$ is large, the diagonal model is a different physics problem wearing our problem’s clothes, and Papers I and II are about an object nobody asked about. If $\Delta$ is small – small enough, in the right sense, relative to the signal we are trying to measure – then the hard quantum disagreement really is governed by the easy classical one, and the scaling laws are theorems about the real thing.

Paper II could not establish which. We could compute $\Delta$ numerically and watch it be small, which is exactly the kind of evidence Scholar would later refuse to accept as proof. What we needed was a bound: a statement that $\Delta$ is, in mean square, $O(d^{-2}d_M^{-1})$ – one power of $d$ below the signal – for reasons written in mathematics, not measured in a simulation.

That bound is the entropy-replacement theorem. Finding it, and proving it without leaving a constant to the machine, was Paper III. The next posts are the routes we took to get there – including the two that did not work.